Sample processing method and device of arrival time estimation model, and electronic equipment

Abstract

The application provides a sample processing method and device of a time of arrival estimation model, and electronic equipment, and the method comprises the following steps: obtaining historical trajectory data of a target object; performing data augmentation processing on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set comprises augmentation processing results of at least two pieces of historical trajectory data; performing feature extraction processing on the historical trajectory data set to obtain a training sample set of the time of arrival estimation model; and performing uniformization processing on the lengths of different training samples in the training sample set to obtain training samples with uniform lengths. Therefore, the accuracy of the samples of the time of arrival estimation model can be effectively improved, so that the result of the time of arrival estimation model is more accurate, and the real-time ETA and future ETA predicted by the time of arrival estimation model will not be fragmented and jump.

Description

[0001] This application claims priority to application number 202111441866.2, filed on November 30, 2021, entitled "Sample Processing Method, Apparatus, and Electronic Equipment for Time of Arrival Prediction Model". Technical Field

[0002] This invention relates to information processing technology, and more particularly to sample processing methods, apparatus, electronic devices, computer program products, and storage media for time of arrival prediction models. Background Technology

[0003] Estimated Time of Arrival (ETA) is an estimate of the travel time for a planned route. ETA is one of the most important metrics for measuring route quality (along with other metrics such as distance, number of traffic lights, number of navigation actions, percentage of main roads, percentage of side roads, congestion rate, and tolls), and it's the primary factor users consider when choosing a route from a set of candidate routes. Currently, the most commonly used algorithms for real-time ETA estimation are rule-based segment-by-segment accumulation methods and tree-based models. However, in practice, the logic for real-time ETA and future ETA is separate and fragmented. This can lead to contradictory results when real-time and future ETA are displayed on the same interface, causing confusion for users and impacting their travel decisions. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a sample processing method, apparatus, electronic device, computer program product and storage medium for a time of arrival (ETA) prediction model, which can effectively improve the accuracy of the samples of the ETA prediction model, thereby making the results of the ETA prediction model more accurate. By using training samples of uniform length, the real-time ETA and future ETA predicted by the ETA prediction model will not be disconnected or jump.

[0005] The technical solution of this invention is implemented as follows: This invention provides a sample processing method for an arrival time estimation model, including: Obtain historical trajectory data of the target object; The historical trajectory data is augmented to obtain a historical trajectory data set, wherein the historical trajectory data set includes the augmentation results of at least two historical trajectory data sets. Feature extraction processing is performed on the historical trajectory data set to obtain the training sample set of the arrival time prediction model; The lengths of different training samples in the training sample set are standardized to obtain training samples of uniform length. This allows the arrival time prediction model to be trained based on these uniformly lengthed training samples, and the arrival time prediction model to predict the arrival time of the target object. This invention also provides a sample processing device for an arrival time estimation model, comprising: The information transmission module is used to acquire historical trajectory data of the target object; An information processing module is used to perform data augmentation processing on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set includes the augmentation processing results of at least two historical trajectory data sets; The information processing module is used to perform feature extraction processing on the historical trajectory data set to obtain a training sample set for the arrival time prediction model. The information processing module is used to unify the lengths of different training samples in the training sample set to obtain training samples of uniform length, so as to train the arrival time prediction model based on the training samples of uniform length, and use the arrival time prediction model to predict the arrival time of the target object.

[0006] In the above scheme, The information processing module is used to extract the departure time, road segment sequence, and actual arrival time based on the historical trajectory data. The information processing module is used to determine the corresponding request time sequence for each historical trajectory data in the historical trajectory data, wherein the request time sequence includes at least two request times; The information processing module is used to construct request data in the form of a quadruple based on each request time, departure time, road segment sequence and actual arrival time in the request time sequence. The information processing module is used to combine the request data corresponding to each request time in the request time sequence with the historical trajectory data to obtain the historical trajectory data set.

[0007] In the above scheme, The information processing module is used to extract road segment sequences from the historical trajectory data set; The information processing module is used to extract basic attribute features, real-time traffic features and historical classic speed features for each road segment in the road segment sequence. The historical classic speed features include at least one of the following: historical classic speed features at the departure time and historical classic speed features at the request time. The information processing module is used to extract the request time feature, departure time feature and actual arrival time feature from the historical trajectory data set; The information processing module is used to perform feature combination processing on the request time feature, departure time feature, actual arrival time feature, basic attribute feature, real-time traffic feature and historical classic speed feature to obtain a training sample set for the arrival time prediction model.

[0008] In the above scheme, The information processing module is used to perform matrix determinant mean processing on the combined results of the basic attribute features, real-time traffic features, and historical classic speed features to obtain training samples of uniform length.

[0009] In the above scheme, The information processing module is used to configure a recurrent neural network for the combined result of the basic attribute features, real-time traffic features, and historical classic speed features. The information processing module is used to iteratively unify the combination of basic attribute features, real-time traffic features and historical classic speed features of each road segment through the recurrent neural network to obtain training samples of uniform length.

[0010] In the above scheme, The information processing module is used to configure a convolutional neural network for the combined result of the basic attribute features, real-time traffic features and historical classic speed features. The information processing module is used to perform convolution processing on the combination of basic attribute features, real-time traffic features and historical classic speed features of each road segment through the convolutional neural network, and to perform average pooling or max pooling processing using the pooling layer of the convolutional neural network to obtain training samples of uniform length.

[0011] This invention also provides an electronic device, the electronic device comprising: Memory, used to store executable instructions; The processor, when running the executable instructions stored in the memory, implements the aforementioned sample processing method for the time-of-arrival prediction model.

[0012] This invention also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implement the aforementioned sample processing method for the time-of-arrival prediction model.

[0013] The embodiments of the present invention have the following beneficial effects: This invention acquires historical trajectory data of a target object; performs data augmentation on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set includes the augmentation results of at least two historical trajectory data sets; performs feature extraction on the historical trajectory data set to obtain a training sample set for an arrival time prediction model; and unifies the length of different training samples in the training sample set to obtain training samples of uniform length. This allows the arrival time prediction model to be trained based on the uniformly lengthed training samples, and the arrival time prediction model to predict the arrival time of the target object. This effectively improves the accuracy of the arrival time prediction model's samples, making the model's results more accurate. The uniformly lengthed training samples ensure that the real-time ETA and future ETA predicted by the arrival time prediction model do not exhibit discrepancies or jumps. Attached Figure Description

[0014] Figure 1 A schematic diagram illustrating a use case of the sample processing method for the arrival time estimation model provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the composition structure of the sample processing device for the arrival time estimation model provided in an embodiment of the present invention; Figure 3 A schematic diagram of an optional sample processing method for the arrival time estimation model provided in this embodiment of the invention; Figure 4 This is a schematic diagram illustrating the data augmentation process for historical trajectory data in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the process of unifying the recurrent neural network in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the process of unifying the recurrent neural network in an embodiment of the present invention; Figure 7 A schematic diagram of an optional process for training the arrival time prediction model provided in an embodiment of the present invention; Figure 8 This is a schematic diagram illustrating the effect of arrival time estimation in an embodiment of the present invention; Figure 9 This is a schematic diagram illustrating the effect of arrival time estimation in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0016] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0017] Before providing a further detailed description of the embodiments of the present invention, the nouns and terms involved in the embodiments of the present invention will be explained, and the nouns and terms involved in the embodiments of the present invention shall be interpreted as follows.

[0018] 1) In response to, used to indicate the conditions or states on which the operation performed depends. When the conditions or states on which it depends are met, one or more operations performed may be performed in real time or with a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations are performed.

[0019] 2) Based on, used to indicate the conditions or states on which the operation is performed depends. When the conditions or states on which it depends are met, one or more operations can be performed in real time or with a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations are performed.

[0020] 3) Model Training: Multi-class classification learning on the image dataset. This model can be built using deep learning frameworks such as TensorFlow and Torch, employing multiple layers of neural networks like CNNs to form a multi-class classification model. The model input is a three-channel or original-channel matrix generated from images read using tools like OpenCV. The model output is the multi-class probability, ultimately outputting the webpage category through algorithms such as softmax. During training, the model gradually approaches the correct trend using objective functions such as cross-entropy until a stopping condition is met, completing the training.

[0021] 4) Neural Network (NN): Artificial Neural Network (ANN), also known as neural network or neural network-like network, is a mathematical or computational model in the fields of machine learning and cognitive science that imitates the structure and function of biological neural networks (the central nervous system of animals, especially the brain) and is used to estimate or approximate functions.

[0022] 5) Intersection travel time: The travel time required to pass through an intersection depends on the location of the traffic lights, the complexity of the intersection, and the time of day (e.g., 8:00-9:00 AM or 1:00-2:00 AM).

[0023] 6) Real-time ETA and Future ETA: Real-time ETA indicates the time required to complete the journey from the current moment. Future ETA indicates the time required to complete the journey from a future moment, such as departing in 15 minutes, departing in 30 minutes, departing at 10:00 tomorrow, departing at 12:00 next Monday, etc. The main difference between real-time ETA and future ETA lies in road conditions. Real-time ETA uses the current road conditions, while future ETA uses estimated road conditions for a future moment. Future ETA can also be referred to as future travel time.

[0024] 7) Request Time and Departure Time: When a user initiates a real-time or future arrival time estimation request on platforms such as apps or mini-programs, the corresponding time is the request time. The departure time is the time the user expects to depart. For real-time ETAs, the request time and departure time are essentially the same; that is, the user departs almost immediately after requesting the ETA. For future ETAs, a user can request the calculation of an ETA departing at 9:00 AM at 8:00 AM, meaning the request time is 8:00 AM and the departure time is 9:00 AM.

[0025] 8) In map applications, a route is simply a complete line connecting the starting point and the destination. In real-world scenarios, the length of a route is usually in the range of one to tens of kilometers.

[0026] 9) Links: In map applications, routes are represented by sequences of links. In map data, roads are divided into line segments, ranging in length from tens of meters to several kilometers. Each line segment is called a link and is assigned a globally unique ID. Therefore, a route on a map is a sequence of all links within that route.

[0027] 10) Actual Time of Arrival (ATA): The actual arrival time of a route can be extracted from the historical data of the map service. Therefore, this data can be used as the ground truth to train machine learning algorithms to predict the arrival time.

[0028] 11) Historical Classic Speed: Based on the historical trajectory of all users, the speed mined for each link. Typically, the cycle is one week, with a granularity of 5 minutes, meaning 7 data points are mined for each link. twenty four 12 = 2016 speed values, representing the travel speed every 5 minutes from Monday to Sunday.

[0029] Before introducing the sample processing method of the arrival time prediction model provided in this application, the prediction process of arrival time in related technologies is first explained. Currently, the most commonly used algorithms for real-time ETA prediction are rule-based segment-by-segment accumulation methods and tree model-based methods. The specific processing steps include: (1) The rule-based segment-by-segment accumulation method relies on human experience. Based on the length, speed, traffic light conditions, etc. of each segment, the passage time of each segment is estimated, and the passage time of each intersection is added together to form the total time of the entire route.

[0030] (2) Tree-based methods no longer estimate each road segment individually. Instead, they first extract features of the entire route, such as the total distance, the average speed at departure, the total number of traffic lights, and the percentage of congested mileage. These features are then input into a tree-based machine learning algorithm for training. The most commonly used algorithm in this category is the Gradient Boosting Decision Tree (GBDT) algorithm. Furthermore, for the prediction of future ETA, typical speeds at the same time period mined from historical trajectories can be used as the real-time speed for future moments.

[0031] However, regardless of which method of estimating arrival time is used, the following defects will occur: (1) The logic of real-time ETA and future ETA is separate and fragmented, which will lead to contradictory results when real-time ETA and future ETA are displayed on the same interface, affecting users' travel decisions; (2) Future ETA only uses the typical speed of the same period in history as the prediction of the real-time speed at the future time, which is a relatively basic strategy and it is not easy to achieve accurate arrival time estimation; (3) Future ETA does not use the real-time traffic conditions and real-time speed at the request time, but in actual use, the traffic conditions and speed at the departure time are also affected by the traffic conditions and speed at the request time, and it is also not easy to achieve accurate arrival time estimation.

[0032] Based on this, embodiments of the present invention provide a sample processing method for an arrival time prediction model, so as to effectively improve the accuracy of the samples of the arrival time prediction model, thereby making the results of the arrival time prediction model more accurate. By using training samples of uniform length, the real-time ETA and future ETA predicted by the arrival time prediction model will not be disconnected or jump.

[0033] Figure 1 This is a schematic diagram illustrating an implementation scenario of the sample processing method for the arrival time estimation model provided in this embodiment of the invention. (See attached diagram.) Figure 1To support an exemplary application, the terminal includes terminal 10-1 and terminal 10-2. The terminal is connected to the sample processing device 200 of the arrival time prediction model via network 300. Network 300 can be a wide area network or a local area network, or a combination of both, and data transmission is achieved using wireless or wired links.

[0034] The terminal (e.g., terminal 10-2) is located on the user side and is used to send an arrival time estimation request. This request seeks an arrival time estimate adapted to the scenario of the target user, who can be various types of map users. For example, arrival time estimation can be a basic function in the map software of terminals 10-1 and 10-2, which allows estimating the time required to complete a given route on the map, given a departure time. This function plays a fundamental and crucial role in many scenarios, including but not limited to: 1. After the user of Terminal 10-1 selects the origin and destination, the map software provides the shortest route. Finding this route requires estimating the time for each candidate route. 2. After the user of Terminal 10-1 initiates navigation, the remaining travel time needs to be continuously reported to the user during navigation. 3. In a food delivery application, Terminal 10-2 needs to allocate orders reasonably to delivery drivers. Allocating orders requires calculating the total time from pickup to delivery based on the customer's location, store location, and delivery driver's location. 4. In a ride-hailing application, Terminal 10-2 needs to match users and taxis reasonably to minimize taxi empty-running time. Route planning requires accurate time estimation for each possible route.

[0035] During this process, the terminals (including terminals 10-1 and 10-2) can obtain relevant historical trajectory data from the corresponding map server 200 via network 300 and present a virtual scene in the display area of ​​the terminal. The sample processing device of the arrival time prediction model set in the terminal can execute the following scheme: The system acquires historical trajectory data of the target object; performs data augmentation on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set includes the augmentation results of at least two historical trajectory data sets; performs feature extraction on the historical trajectory data set to obtain a training sample set for the arrival time prediction model; and unifies the lengths of different training samples in the training sample set to obtain training samples of uniform length, so as to train the arrival time prediction model based on the training samples of uniform length and use the arrival time prediction model to predict the arrival time of the target object.

[0036] In some embodiments, terminal 10-1 may also install and run applications that support virtual scenes. This application can be a virtual reality application, a 3D map application, a first-person shooter (FPS) game, a multiplayer online battle arena (MOBA) game, or other virtual scenarios. Taking a shooting game as an example, users can control virtual objects to perform actions at different locations within the estimated arrival time, or to perform actions at different points in time while moving along the estimated arrival time. For example, at different locations within the estimated arrival time, users can perform actions such as freefalling, gliding, or parachuting in the sky. They can also perform actions such as running, jumping, crawling, or bending forward at different locations within the estimated arrival time on land. Users can also control virtual objects to swim, float, or dive in the ocean at different locations within the estimated arrival time. Of course, users can also control virtual objects to move in the virtual scenario using virtual vehicles, with their real-time location displayed at different locations within the estimated arrival time. For example, the virtual vehicle could be a virtual car, a virtual aircraft, or a virtual yacht moving along the estimated arrival time. These examples are only provided for illustration, and this embodiment of the invention does not impose any specific limitations on them. Users can also control virtual objects to interact with other virtual objects through virtual props, such as fighting. This invention does not specifically limit the type of virtual props. When controlling virtual objects to perform actions at different positions in the estimated arrival time, the sample processing method of the arrival time estimation model provided in this application can be used to process the historical trajectory data in the virtual scene, and the arrival time of the virtual object to a certain position can be predicted by the arrival time estimation model, or the arrival time of the virtual object to the attack position of any virtual prop can be predicted by the arrival time estimation model. This invention does not limit this.

[0037] The following describes a sample processing apparatus for the arrival time estimation model used in implementing the sample processing method for the arrival time estimation model of the embodiments of the present invention. The sample processing apparatus for the arrival time estimation model can be implemented in various forms, such as a terminal with sample processing functionality for the arrival time estimation model, or a server equipped with sample processing functionality for the arrival time estimation model, for example, the preceding... Figure 1 Server 200 in the middle. Figure 2 This is a schematic diagram of the composition of the sample processing device for the arrival time prediction model provided in this embodiment of the invention. It can be understood that... Figure 2 This is only an exemplary structure of the sample processing device for the arrival time estimation model, and not the entire structure; it can be implemented as needed. Figure 2 The structure shown may be part or all of the structure.

[0038] The sample processing device for the arrival time estimation model provided in this embodiment of the invention includes: at least one processor 201, a memory 202, a user interface 203, and at least one network interface 204. The various components in the sample processing device for the arrival time estimation model are coupled together via a bus system 205. It can be understood that the bus system 205 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 205 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 2 The general designated all buses as Bus System 205.

[0039] The user interface 203 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0040] It is understood that memory 202 can be volatile memory or non-volatile memory, or both. In this embodiment of the invention, memory 202 is capable of storing data to support the operation of a terminal (such as 10-1). Examples of this data include any computer programs used to operate on the terminal (such as 10-1), such as operating systems and applications. The operating system includes various system programs, such as the framework layer, core library layer, driver layer, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications.

[0041] In some embodiments, the sample processing device for the arrival time estimation model provided in this invention can be implemented using a combination of hardware and software. For example, the sample processing device for the arrival time estimation model provided in this invention can be a processor in the form of a hardware decoding processor, programmed to execute the sample processing method for the arrival time estimation model provided in this invention. For instance, the processor in the form of a hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0042] As an example of the hardware and software combined implementation of the sample processing device for the arrival time estimation model provided in this embodiment of the invention, the sample processing device for the arrival time estimation model provided in this embodiment of the invention can be directly embodied as a combination of software modules executed by the processor 201. The software modules can be located in a storage medium, which is located in the memory 202. The processor 201 reads the executable instructions included in the software modules in the memory 202 and combines them with the necessary hardware (e.g., including the processor 201 and other components connected to the bus 205) to complete the sample processing method for the arrival time estimation model provided in this embodiment of the invention.

[0043] As an example, processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0044] As an example of the hardware implementation of the sample processing device for the arrival time prediction model provided in this embodiment of the invention, the device provided in this embodiment of the invention can be directly executed by a processor 201 in the form of a hardware decoding processor. For example, it can be executed by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the sample processing method for the arrival time prediction model provided in this embodiment of the invention.

[0045] In this embodiment of the invention, memory 202 is used to store various types of data to support the operation of the sample processing apparatus for the time-of-arrival estimation model. Examples of such data include any executable instructions for operation on the sample processing apparatus for the time-of-arrival estimation model, such as executable instructions that can be included in a program implementing the sample processing method for the time-of-arrival estimation model according to this embodiment of the invention.

[0046] In other embodiments, the sample processing device for the arrival time estimation model provided in this invention can be implemented in software. Figure 2A sample processing device for an arrival time estimation model stored in memory 202 is shown. This device can be software in the form of programs and plug-ins, and includes a series of modules. As an example of a program stored in memory 202, it may include a sample processing device for an arrival time estimation model. The sample processing device for the arrival time estimation model includes the following software modules: The information transmission module 2081 is used to acquire historical trajectory data of the target object.

[0047] The information processing module 2082 is used to perform data augmentation processing on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set includes the augmentation processing results of at least two historical trajectory data.

[0048] The information processing module 2082 is used to perform feature extraction processing on the historical trajectory data set to obtain a training sample set for the arrival time prediction model.

[0049] The information processing module 2082 is used to unify the lengths of different training samples in the training sample set to obtain training samples with uniform lengths, so as to train the arrival time prediction model based on the training samples with uniform lengths, and use the arrival time prediction model to predict the arrival time of the target object.

[0050] In some embodiments, the sample processing device 200 for the time-of-arrival estimation model can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal (such as terminal 10-1) can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment of the invention.

[0051] according to Figure 2In one aspect of this application, the sample processing apparatus for the time-of-arrival (TOA) estimation model also provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform various embodiments and combinations thereof provided in the various optional implementations of the sample processing method for the TOA estimation model described above.

[0052] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0053] Continue to combine Figure 2 The sample processing apparatus for the arrival time estimation model shown illustrates the sample processing method for the arrival time estimation model provided in this embodiment of the invention. See also... Figure 3 , Figure 3 This is an optional flowchart illustrating the sample processing method for the arrival time estimation model provided in this embodiment of the invention. It can be implemented by running on... Figure 1 The software or component with arrival time prediction function in terminal 10-1 or 10-2 shown executes the sample processing method of the arrival time prediction model provided in this application. This ensures the accuracy of the arrival time prediction model's samples while making the model's results more accurate. Furthermore, by using training samples of uniform length, the real-time ETA predicted by the arrival time prediction model and the future ETA do not exhibit any disconnects or jumps. Figure 3 The steps shown can be performed by various electronic devices of a sample processing device that runs an arrival time prediction model. For example, it can be a terminal such as a motion-sensing game console, or the sample processing method of the arrival time prediction model provided in this application can be performed by software or functional modules with arrival time prediction function in a mobile phone.

[0054] The following is combined with Figure 3 The steps shown are illustrated using the sample processing device for the arrival time estimation model to implement the sample processing method for the arrival time estimation model provided in this embodiment of the invention.

[0055] Step 301: The sample processing device of the arrival time prediction model acquires the historical trajectory data of the target object.

[0056] Wherein, one piece of trajectory data may include roads with the same attribute, and may also include multiple roads with different attributes. For example, a certain piece of trajectory data includes 2 roads with different attributes, one of which is an unobstructed road and the other is a congested road. For another example, a certain piece of trajectory data includes 3 roads with different attributes, one of which is an unobstructed road with 3 lanes, another is an unobstructed road with 2 lanes, and the other is a congested road with 2 lanes.

[0057] Step 302: a sample processing apparatus for an arrival time estimation model performs data amplification processing on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set includes amplification processing results of at least two pieces of historical trajectory data.

[0058] In some embodiments of the present invention, performing data amplification processing on the historical trajectory data to obtain a historical trajectory data set may be implemented in the following manner: Based on the historical trajectory data, extract the departure time, road segment sequence and actual arrival time; for each piece of historical trajectory data in the historical trajectory data, determine a corresponding request time sequence, wherein the request time sequence includes at least two request times; construct request data in a quadruple format based on each request time in the request time sequence, the departure time, the road segment sequence and the actual arrival time; combine the request data corresponding to each request time in the request time sequence with the historical trajectory data to obtain the historical trajectory data set. Wherein, with reference to Figure 4 , Figure 4 is a schematic diagram of data amplification processing on historical trajectory data in an embodiment of the present invention. For a collected trajectory, its departure time t0 and actual arrival time are determined, and the request time t req (t req < t0) can be arbitrary, so any number of arrival time estimation requests can be simulated, and they correspond to the same actual arrival time. In this way, any number of request data can be amplified based on one trajectory.

[0059] Therefore, firstly, the departure time t0, road segment sequence and actual arrival time corresponding to each historical trajectory are extracted from the historical trajectory data; then, for each historical trajectory, a request time sequence is calculated, the request time sequence includes at least two request times, and when calculating the request time, refer to Formula 1: t req = t0 - k, and k = 10, 20, 30..., the unit is minutes Formula 1 After the request time sequence is obtained through calculation by Formula 1, for each historical trajectory and each request time, t reqThe time to arrival (t0), the route sequence, and the actual arrival time can form a quadruple, and each quadruple represents a new request data entry. Finally, by combining all historical trajectories and all request times, the extracted request data constitutes the augmented dataset. Figure 4 As shown, by performing data augmentation on the historical trajectory data, the resulting historical trajectory data set includes four different historical trajectory data points, namely 8:00 (t req ) Request ETA departing at 10:00 (t0) and 8:30 (t1) req ) Request ETA departing at 10:00 (t0) and 9:00 (t... req ) Request an ETA departing at 10:00 (t0) and at 9:30 (t... req ) Request an ETA departing at 10:00 (t0).

[0060] Step 303: The sample processing device of the arrival time prediction model performs feature extraction processing on the historical trajectory data set to obtain the training sample set of the arrival time prediction model.

[0061] In some embodiments of the present invention, feature extraction processing is performed on the historical trajectory data set to obtain a training sample set for the arrival time prediction model, which can be achieved in the following ways: Extract road segment sequences from the historical trajectory data set; extract basic attribute features, real-time traffic features, and historical classic speed features for each road segment in the road segment sequence, wherein the historical classic speed includes at least one of the following: historical classic speed features at the departure time and historical classic speed features at the request time; extract request time features, departure time features, and actual arrival time features from the historical trajectory data set; perform feature combination processing on the request time features, departure time features, actual arrival time features, basic attribute features, real-time traffic features, and historical classic speed features to obtain the training sample set for the arrival time prediction model. For the constructed four-tuple format request data, the request time t... req The departure time t0 and the departure time can be two features, for example, the time difference t between them. 0- t req It can also be used as a feature.

[0062] For a road segment sequence, since it is a sequence (including information from multiple road segments), it is necessary to extract the following four types of features for each road segment: 1) Basic attribute characteristics: such as road segment mileage, road grade, width, whether there are traffic lights, etc.

[0063] 2) Near real-time traffic features at the requested time treq: For example, within the time interval from treq-30min to treq, with a granularity of 5min, the features corresponding to the 6 traffic status features, 6 real-time speed features, and 6 real-time traffic flow features of the road segment.

[0064] 3) Historical classic speed characteristics at departure time t0: For example, within the time interval from t0-30min to t0, with a granularity of 5min, the characteristics corresponding to the 6 historical classic speed characteristics and the standard deviation of the 6 historical speeds.

[0065] 4) In some embodiments of the present invention, optionally, the request time t can also be obtained. req Historical classic speed characteristics: for example, t req Within the time interval from -30min to t0, with a granularity of 5min, the six historical classic speed characteristics and the standard deviation characteristics of the six historical speeds of the road segment are analyzed.

[0066] Through the above processing, a K-dimensional vector is obtained for each road segment. The K-dimensional vectors of all road segments form a sequence of length M. Therefore, the feature extracted from the road segment sequence is an M-dimensional vector. The matrix of K; finally, by combining the features of the requested time, departure time, actual arrival time, basic attribute features, real-time traffic conditions, and historical speed features, a training sample set for the arrival time prediction model can be obtained.

[0067] Step 304: The sample processing device of the arrival time prediction model unifies the length of different training samples in the training sample set to obtain training samples with uniform length, so as to train the arrival time prediction model based on the training samples with uniform length, and use the arrival time prediction model to predict the arrival time of the target object.

[0068] In some embodiments of the present invention, the lengths of different training samples in the training sample set are standardized to obtain training samples of uniform length. This can be achieved in the following ways: The combined results of the basic attribute features, real-time traffic features, and historical classic speed features are subjected to matrix determinant mean processing to obtain training samples of uniform length. Among them, since it includes features extracted from the link sequence, it is an M... The matrix M is K, and the number of links in each route varies, so M is an uncertain value. Therefore, for K-dimensional vector sequences of variable length, a unified processing is needed, which can be achieved by directly processing M. The mean of all rows in a K-dimensional matrix is ​​calculated to obtain a K-dimensional vector. This K-dimensional vector, along with the request time feature, departure time feature, and actual arrival time feature, can be combined into a set of fixed-length features. This set of features is then input into an ETA model (for example, a mature machine learning model such as a linear model or a deep model can be used for training), resulting in a unified ETA model. Thus, by using training samples of uniform length, the real-time ETA predicted by the arrival time estimation model and the future ETA will not produce any discrepancies or jumps.

[0069] In some embodiments of the present invention, the lengths of different training samples in the training sample set are standardized to obtain training samples of uniform length. This can be achieved in the following ways: A recurrent neural network is configured for the combined results of the basic attribute features, real-time traffic features, and historical classic speed features. The recurrent neural network iteratively unifies the combination of basic attribute features, real-time traffic features, and historical classic speed features for each road segment, resulting in training samples of uniform length. Among these, reference... Figure 5 , Figure 5 This is a schematic diagram illustrating the unified processing process of the recurrent neural network in this embodiment of the invention. The recurrent neural network processes (e.g., using LSTM, GRU, etc.) the features corresponding to one road segment at each time step until all road segment features have been processed, outputting a fixed-length K-dimensional vector. This K-dimensional vector, together with the request time feature, departure time feature, and actual arrival time feature, can be combined into a set of fixed-length features, which are then input into the ETA model (e.g., the ETA model can use mature machine learning models such as linear models or deep models) for training, thereby obtaining a unified ETA model. Thus, by using training samples of uniform length, the real-time ETA predicted by the arrival time prediction model and the future ETA will not produce any disconnect or jumps.

[0070] In some embodiments of the present invention, the lengths of different training samples in the training sample set are standardized to obtain training samples of uniform length. This can be achieved in the following ways: A convolutional neural network is configured for the combined results of the basic attribute features, real-time traffic features, and historical classic speed features. The convolutional neural network performs convolution processing on the combination of basic attribute features, real-time traffic features, and historical classic speed features for each road segment, and then uses the pooling layers of the convolutional neural network to perform average pooling or max pooling to obtain training samples of uniform length. Among these, reference... Figure 6 , Figure 6 This is a schematic diagram illustrating the process of unifying the convolutional neural network in an embodiment of the present invention. One-dimensional convolution operations are used to process M... The K-matrix is ​​processed through convolutional layers, followed by average pooling or max pooling to obtain a fixed-length K-dimensional vector. This K-dimensional vector, along with the request time feature, departure time feature, and actual arrival time feature, can be combined into a set of fixed-length features. This set of features is then input into the ETA model (for example, the ETA model can use mature machine learning models such as linear models or deep models) for training, thereby obtaining a unified ETA model. Thus, by using training samples of uniform length, the real-time ETA predicted by the arrival time prediction model and the future ETA will not produce any disconnects or jumps.

[0071] The sample processing method for the arrival time estimation model provided in this application is based on artificial intelligence. The technical solution provided by this invention utilizes artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can react in a way similar to human intelligence. Artificial intelligence studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making functions.

[0072] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, and inductive learning. With the research and advancement of artificial intelligence technology, it is being researched and applied in multiple fields. For example, in this application, when processing data in virtual scenarios using artificial intelligence, the arrival time prediction model can determine the corresponding arrival time prediction through different neural networks and processing processes, enabling users with different needs to obtain accurate arrival time predictions and a better user experience. Figure 2As shown, the sample processing device 200 for the arrival time prediction model is used to implement the sample processing method of the arrival time prediction model provided in this application through the deployed arrival time prediction model after receiving the sent route generation request. As an example, the sample processing device 200 for the arrival time prediction model is used to implement the sample processing method of the arrival time prediction model provided in this invention. The trained arrival time prediction model can be stored in the storage medium of the sample processing device 200 for the purpose of generating different arrival time prediction model samples in different scenarios (such as virtual scene shooting games, multiplayer online battle arena (MOBA), racing games (RCG), and sports games (SPG)) and environments that require arrival time prediction, such as online ride-hailing and food delivery). This enables the training of the arrival time prediction model based on the uniformly long training samples, and the prediction of the arrival time of the target object using the arrival time prediction model.

[0073] Of course, the sample processing device for the arrival time estimation model provided by this invention can be trained based on the arrival time estimation value generated in the environment of the same type of virtual scene, or it can be trained and adjusted according to different levels of the target object. Finally, the arrival time estimation value that is adapted to the virtual scene by the arrival time estimation model is presented on the user interface. The arrival time estimation model and the arrival time estimation value adapted to the virtual scene can also be called by other applications (such as game emulators or motion-sensing game devices). Of course, the arrival time estimation model that matches different types of games can also be transferred to online FPS games with instant messaging process or offline FPS games and cloud games. Alternatively, the arrival time estimation in map software can be called by ride-hailing software.

[0074] In some embodiments, before deploying the time-of-arrival (TOA) prediction model, it is necessary to train the TOA model to determine the parameters of different neural networks within the TOA model, in order to accurately process map location coordinates and map type parameters. Figure 7 This is an optional flowchart illustrating the time-of-arrival prediction model training method provided in this embodiment of the invention. It can be understood that... Figure 7The steps shown can be performed by various electronic devices that run the arrival time prediction model training device, such as software devices with an arrival time prediction model training device. The trained arrival time prediction model is then encapsulated in corresponding software or functional modules with arrival time prediction functions to provide services for processing arrival time prediction values ​​for users with different needs.

[0075] Step 701: The arrival time prediction model training device determines the historical parameters of the target object based on the type of the virtual scene in which the target object is located.

[0076] Step 702: The arrival time prediction model training device determines training samples of uniform length that match the arrival time prediction model based on the historical parameters of the target object.

[0077] The uniformly long training samples include at least one set of training samples. After obtaining the training samples, information annotation is required on the obtained training samples.

[0078] Step 703: The arrival time prediction model training device extracts different training samples from the training samples of uniform length based on the noise threshold matched by the arrival time prediction model to form an actual training sample set.

[0079] In some embodiments of the present invention, when the virtual scene in which the target object is located is a role-playing game, a dynamic noise threshold matching the usage environment of the arrival time prediction model is determined; noise is removed from training samples of uniform length based on the dynamic noise threshold to form an actual training sample set matching the dynamic noise threshold; when the virtual scene in which the target object is located is a battle game, a fixed noise threshold corresponding to the arrival time prediction model is determined, and noise is removed from training samples of uniform length based on the fixed noise threshold to form an actual training sample set matching the fixed noise threshold. Because the virtual scenarios of the arrival time estimation models differ, the dynamic noise thresholds that match the usage environment of the models also differ. For example, role-playing games can be played online via an instant messaging client process or offline via an offline FPS game process. The complexity of online FPS games is usually greater than that of offline FPS games. Therefore, the dynamic noise threshold matching the usage environment of the arrival time estimation model needs to be lower than the dynamic noise threshold in the usage environment of role-playing games played offline by users with different needs. Training samples exceeding the noise threshold are deleted. By using different dynamic noise thresholds, different types of games can be adapted to effectively filter training samples, so that users can obtain better arrival time estimates from the deployed and trained arrival time estimation models.

[0080] In some embodiments of the present invention, reference is made to Figure 8 , Figure 8 This is a schematic diagram illustrating the effect of arrival time prediction in an embodiment of the present invention. When the virtual scene in which the target object is located is a navigation environment, a fixed noise threshold corresponding to the arrival time prediction model is determined. Noise is removed from training samples of uniform length based on the fixed noise threshold to form an actual training sample set matching the fixed noise threshold. For the arrival time prediction model deployed in the navigation software server, the fixed noise threshold effectively improves the speed and accuracy of training sample acquisition, reducing the waiting time for users with different needs. Through the arrival time prediction model in the navigation software server, such as... Figure 8 As shown, when initiating navigation, the server first provides several candidate routes, then uses an arrival time prediction model trained with a fixed noise threshold to calculate the estimated arrival time of each candidate route, and finally selects the fastest route to provide to the user. After entering navigation mode, the remaining travel time is calculated at regular intervals using this invention, making it convenient for the user to plan their trip.

[0081] refer to Figure 9, Figure 9 This is a schematic diagram illustrating the effect of arrival time prediction in an embodiment of the present invention. The estimated arrival time of each candidate route is calculated using an arrival time prediction model trained with a fixed noise threshold. At the same time, the ETA of each route can be used to evaluate the merits of each route, thereby pushing the optimal route to the user.

[0082] Step 704: The arrival time prediction model training device trains the arrival time prediction model based on the extracted actual training sample set, and determines the model parameters of the feature extractor network and the model parameters of the sample classifier network in the arrival time prediction model.

[0083] Step 705: After determining the network parameters of different neural networks in the arrival time prediction model, the arrival time prediction model training device deploys the trained arrival time prediction model in the corresponding game terminal.

[0084] Beneficial technical effects: This invention acquires historical trajectory data of a target object; performs data augmentation on the historical trajectory data to obtain a historical trajectory data set, wherein the historical trajectory data set includes the augmentation results of at least two historical trajectory data sets; performs feature extraction on the historical trajectory data set to obtain a training sample set for an arrival time prediction model; and unifies the length of different training samples in the training sample set to obtain training samples of uniform length. This allows the arrival time prediction model to be trained based on the uniformly lengthed training samples, and the arrival time prediction model to predict the arrival time of the target object. This effectively improves the accuracy of the arrival time prediction model's samples, making the model's results more accurate. The uniformly lengthed training samples ensure that the real-time ETA and future ETA predicted by the arrival time prediction model do not exhibit discrepancies or jumps.

[0085] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A sample processing method for an arrival time prediction model, characterized in that, The method includes: Obtain historical trajectory data of the target object; Based on the historical trajectory data, the departure time, route sequence, and actual arrival time are extracted. For each historical trajectory data in the historical trajectory data, a corresponding request time sequence is determined, wherein the request time sequence includes at least two request times; Based on each request time, departure time, route sequence, and actual arrival time in the request time sequence, construct request data in the format of a quadruple; The request data corresponding to each request time in the request time sequence is combined with the historical trajectory data to obtain a historical trajectory data set. Extract the road segment sequence from the historical trajectory data set; Extract the basic attribute features, real-time traffic features, and historical classic speed features for each road segment in the road segment sequence. The historical classic speed features include at least one of the following: historical classic speed features at the departure time and historical classic speed features at the request time. Extract the request time feature, departure time feature, and actual arrival time feature from the historical trajectory data set; The request time feature, departure time feature, actual arrival time feature, basic attribute feature, real-time traffic feature and historical classic speed feature are combined to obtain the training sample set of the arrival time prediction model. The lengths of different training samples in the training sample set are standardized to obtain training samples of uniform length. The arrival time prediction model is then trained based on the uniformly lengthed training samples, and the arrival time prediction model is used to predict the arrival time of the target object.

2. The method according to claim 1, characterized in that, The process of unifying the lengths of different training samples in the training sample set to obtain training samples of uniform length includes: The combined results of the basic attribute features, real-time traffic features, and historical classic speed features are averaged in each column of the matrix to obtain training samples of uniform length.

3. The method according to claim 1, characterized in that, The process of unifying the lengths of different training samples in the training sample set to obtain training samples of uniform length includes: Configure a recurrent neural network for the combined result of the basic attribute features, real-time traffic features, and historical classic speed features; The recurrent neural network iteratively unifies the combination of basic attribute features, real-time traffic features, and historical speed features for each road segment to obtain training samples of uniform length.

4. The method according to claim 1, characterized in that, The process of unifying the lengths of different training samples in the training sample set to obtain training samples of uniform length includes: Configure a convolutional neural network for the combined result of the basic attribute features, real-time traffic features, and historical classic speed features; The convolutional neural network performs convolution processing on the combination of basic attribute features, real-time traffic features, and historical classic speed features of each road segment, and uses the pooling layer of the convolutional neural network to perform average pooling or max pooling to obtain training samples of uniform length.

5. A sample processing device for an arrival time prediction model, characterized in that, The device includes: The information transmission module is used to acquire historical trajectory data of the target object; The information processing module is used to extract the departure time, road segment sequence, and actual arrival time based on the historical trajectory data. For each historical trajectory data in the historical trajectory data, a corresponding request time sequence is determined, wherein the request time sequence includes at least two request times; Based on each request time, departure time, route sequence, and actual arrival time in the request time sequence, construct request data in the format of a quadruple; The request data corresponding to each request time in the request time sequence is combined with the historical trajectory data to obtain a historical trajectory data set. The information processing module is also used to extract road segment sequences from the historical trajectory data set; Extract the basic attribute features, real-time traffic features, and historical classic speed features for each road segment in the road segment sequence. The historical classic speed features include at least one of the following: historical classic speed features at the departure time and historical classic speed features at the request time. Extract the request time feature, departure time feature, and actual arrival time feature from the historical trajectory data set; The request time feature, departure time feature, actual arrival time feature, basic attribute feature, real-time traffic feature and historical classic speed feature are combined to obtain the training sample set of the arrival time prediction model. The information processing module is also used to unify the lengths of different training samples in the training sample set to obtain training samples with uniform lengths, so as to train the arrival time prediction model based on the training samples with uniform lengths, and use the arrival time prediction model to predict the arrival time of the target object.

6. The apparatus according to claim 5, characterized in that, The information processing module is also used for: The combined results of the basic attribute features, real-time traffic features, and historical classic speed features are averaged in each column of the matrix to obtain training samples of uniform length.

7. The apparatus according to claim 5, characterized in that, The information processing module is also used for: Configure a recurrent neural network for the combined result of the basic attribute features, real-time traffic features, and historical classic speed features; The recurrent neural network iteratively unifies the combination of basic attribute features, real-time traffic features, and historical speed features for each road segment to obtain training samples of uniform length.

8. The apparatus according to claim 5, characterized in that, The information processing module is also used for: Configure a convolutional neural network for the combined result of the basic attribute features, real-time traffic features, and historical classic speed features; The convolutional neural network performs convolution processing on the combination of basic attribute features, real-time traffic features, and historical classic speed features of each road segment, and uses the pooling layer of the convolutional neural network to perform average pooling or max pooling to obtain training samples of uniform length.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; A processor, when running executable instructions stored in the memory, implements the sample processing method of the arrival time estimation model according to any one of claims 1 to 4.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the sample processing method of the arrival time prediction model according to any one of claims 1 to 4.

11. A computer-readable storage medium storing executable instructions, characterized in that, When the executable instructions are executed by the processor, they implement the sample processing method of the arrival time prediction model according to any one of claims 1 to 4.

Images